The exponential leap in generative AI is already transforming many industries: optimizing workflows, helping human teams focus on value added tasks and accelerating time to market. Life sciences industry is beginning to take notice and aims to leapfrog the technological advances. Life sciences industry has—for decades now—moved from the traditional discovery-based drug development to target market-based drug development paradigm. Yet, it is burdened by long R&D cycles and labor-intensive clinical, manufacturing and compliancy regimens.
The industry is under tremendous pressure to accelerate drug development at an optimal cost, automate time- and labor-intensive tasks like document or report creation to preserve employee morale, and accelerate delivery. With BioPharma and Medical Device organizations increasingly adopting digital transformation and engagement strategies—combined with the paradigm shift brought about by the Covid19 pandemic—the industry is experiencing an explosion of digital data being created in the commercial, supply chain, clinical and pharmacovigilance areas of the value chain, and as well as in other enterprise business functions.
This digital data is coming at the industry in various formats, like unstructured text, images, PDFs and emails. The explosion in digital data—in combination with declining availability of skilled and willing human resources to ingest and process the digital data in a compliant manner—is forcing life sciences organizations to explore AI, machine learning and now generative AI technologies. Some examples of potential use cases for generative AI in life sciences include but are not limited to:
AI for Medical Legal Review (MLR): Increasing globalization and exponential growth in digital marketing techniques has been putting strain on the already complex, time consuming and challenging process. generative AI has the potential to process digital content at scale and produce an effective MLR output, which can then be leveraged by the human marketing team, accelerating and simplifying the process.
AI for generating Clinical Study reports (CSR): Generative AI has the potential to create a “first attempt” report, which can offset 80% of human effort, accelerating the process, bringing in consistency and freeing up valuable bandwidth for other high value tasks.
Adverse Event (AE) Narrative generation: This highly regulated, time-consuming task of generating an adverse event narrative requires highly regulated business functions and highly skilled roles within life sciences organizations and require coordination of manual, sometimes tedious, tasks that can produce potentially inaccurate or inconsistent results. Leveraging generative AI to augment human team capabilities presents an opportunity for Clients to reduce costs by 30%-50%, while accelerating time to market related to this process by at least 50% and improving scalability, quality, and consistency of generated reports.
Accelerate mRNA medicines design: Moderna, which has been leveraging machine learning and AI to advance the field of messenger RNA (mRNA) to create a diverse clinical portfolio of vaccines and therapeutics across seven modalities, is partnering with IBM to leverage generative AI to design mRNA medicines with optimal safety and performance.
Other use cases where generative AI models can help life sciences organizations unleash competitive advantage are:
Research & Development: Drug discovery & development, quality content creation and review, quality and regulatory intelligence, AE Narrative Generation, intelligent submissions, synthetic data generation.
Commercial: Marketing content creation, patient experience, rep onboarding & training sales enablement and knowledge hub.
Human Resources: Create cob descriptions, skill requirements, create interview questions from a job description, assess candidates against a job spec, learning & teaching assistant, quiz creation, content creation and more.
Manufacturing: Quality control and inspection, operator / lab tech training conversational search through SOP’s, content creation and more.
Summarization: call center interactions, documents such as financial reports, analyst articles, emails, news, media trends and more.
Conversational Knowledge: Reviews, knowledge base, product descriptions and more.
Content creation: Personas, user stories, synthetic data, generating images, personalized UI, marketing copy, email and social responses and more.
Code creation: Code co-pilot, code conversion, create technical documentation, test cases and more.
We believe that leveraging generative AI-Automation can drive benefits in life sciences—including in regulated domains—and reduce cycle times for creating AE Narratives by at least 50%, based on work being done by IBM Consulting and the Pharmacovigilance group at a global BioPharma company.
In this blog post, we will showcase how IBM Consulting is partnering with AWS and leveraging Large Language Models (LLMs), on IBM Consulting’s generative AI-Automation platform (ATOM), to create industry-aware, life sciences domain-trained foundation models to generate first drafts of the narrative documents, with an aim to assist human teams.
Why IBM Consulting for generative AI on AWS?
For more than a decade, IBM Consulting has helped clients drive value through AI, machine learning and automation solutions to optimize business process and IT operations across industries. More recently, IBM Consulting has been partnering with enterprises to deploy foundation models to reimagine core workflows and realize value—reducing costs, turnaround time, and improving productivity and is committed to helping enterprises navigate and unlock value from the seismic changes driven by AI. With that in mind, IBM Consulting recently announced a generative AI Center of Excellence with 1000+ consultants skilled in generative AI and accelerator toolkits purpose-built for foundation models and LLMs; through this, IBM Consulting is helping enterprises develop and deploy production-grade generative AI models.
IBM is a Premier Consulting Partner for AWS with 20K+ AWS certified professionals across the globe, 16 service validations and 16 AWS competencies, becoming the fastest Global GSI to secure more AWS competencies and certifications among top-16 AWS Premier GSI’s within 18 months. At re:Invent 2022, IBM Consulting was awarded the Global Innovation Partner of the Year and the GSI Partner of the Year for Latin America, cementing client and AWS trust in IBM Consulting as a partner of choice when it comes to AWS.
In the AI domain, IBM has 21K+ data Scientists, AI Engineers, and consultants and has executed 40K+ AI and analytics engagements. But with great power comes great responsibility, and this is especially true for generative AI. IBM Consulting has been driving a responsible and ethical approach to AI for more than five years now, mainly focused on these five basic principles:
Explainability: How an AI model arrives at a decision should be able to be understood, with human-in-the-loop systems adding more credibility and help mitigating compliance risks.
Fairness: AI models should treat all groups equitably.
Robustness: AI systems should be able to withstand attacks to the training data.
Transparency: All relevant aspects of an AI system should be available to the public for evaluation.
Privacy: The data used in AI systems should be secure, and when that data belongs to an individual, the individual should understand how it is being used.
To help life sciences organizations follow GxP guidelines and regulations when developing or manufacturing drugs and medical devices, IBM Consulting leverages its vast GxP experience and AWS best practices around GxP, HIPAA and other compliance programs to deliver compliant, regulated, validated and secure solutions.
How to build a generative AI pipeline in AWS for narrative generation?
Currently, creating narratives for adverse events is an intensive manual process in healthcare. When an adverse event is reported, clinical and safety teams manually read and process several details—patient current and historical health and medical information, the event data and more—and manually write a detailed report, as is needed by the regulatory authorities. With the advent of generative AI, we believe these processes can be augmented to free up capacity for clinical and safety teams to shift to higher value tasks such as reviewing the narratives as well as enabling the teams to focus on more complex tasks.
We explored multiple options for the task of generating adverse event narratives using generative AI. Ultimately, one of the HuggingFace Large Language Models on Amazon Sagemaker JumpStart was selected to build the Adverse event narratives for multiple reasons: it has a permissive license that allows commercial usage, clear model/data cards for the source model that can explain its data lineage, the ability to fine-tune the model within Sagemaker Jumpstart, and robust capability to generate adverse event narrative text with minimal amount of fine-tuning.
The high-level pipeline for this process is shown in Figure 1. We started with prepping the proprietary structured data to clean and make it ready in a format to be able to pass within prompts for fine-tuning and inferencing. The Large Language Model was then fine-tuned in Amazon Sagemaker on a training dataset of 500+ records that describes patient health information, adverse events and medical information, using the pipeline shown below. Amazon Sagemaker is an optimal platform for generative AI owing to several in built functionalities (ability to select models from a catalog, no code approach to train models, functionalities to set up additional pipelines and monitor.) Once fine tuned, the deployed model was used to inference on a test data to create the AE narratives (see Figure 2 for a sample). Additionally, the team of Safety and Clinical Subject Matter Experts validated the narrative generation using ground truth documents and manually analyzed them to ensure that the generative AI-Automation pipeline was reliable and not subject to hallucinations.
In addition to this, IBM Consulting recently launched watsonx.data on AWS, an open, hybrid, governed data store to help enterprises scale analytics and AI. IBM Consulting is also partnering with AWS to integrate the upcoming Amazon Bedrock, a fully managed service that makes FMs from leading AI startups and Amazon available via an API, into ATOM, to help clients build and scale generative AI use cases, while strengthening cybersecurity and compliance.
As per FAERS database, the number of reported AEs has grown 2.5x in 10 years, from 2012 to 2022. Regardless of volumes, companies must report these events rapidly to regulators and act quickly on safety signals. The burden from growing event volumes is reflected in budgets that are expected to grow from an estimated USD 4 billion in 2017 to over 6 billion by 2020.
According to a top 10 major US based life sciences client that IBM consulting is currently working with, leveraging generative AI in a compliant and responsible fashion has the potential to reduce the manual labor for creating AE reports by 50%. Combining that with an AI driven, human in the loop, language translation solution, can further optimize operation costs and free up valuable human teams to focus on value added tasks.
In a nod to the growing usage of Machine learning in life sciences, FDA has now cleared more than 500 medical algorithms that are commercially available in the United States. More than half of algorithms on the U.S. market were cleared between 2019 to 2022, with more than 300 apps in just four years. In October 2022 alone, the FDA approved 178 new AI/ML systems, a number expected to grow rapidly into the future.
This momentum creates an enormous business value for life sciences clients looking to innovate across the value chain, leveraging cutting edge technologies like generative AI.
How IBM Consulting can support clients on their journey to leveraging Foundation Models?
IBM Consulting has the expertise and experience to support clients with varying degrees of maturity on their generative AI journey. On a high level, IBM Consulting leverages the following pillars to meet clients where they are:
Generative AI Strategy and Center of Excellence setup: Standardized consulting engagement to inform, engage, discover and assess new use cases for foundation models.
Foundation Model Hackathon: A 2-day hackathon to ideate and prototype innovative AI solutions for specific use case domains—leveraging standard cloud APIs or open-source foundation models (GPT, BERT and others).
Jumpstart for foundation model: Leverage IBM Garage to jumpstart the use of foundation models and implement proven IBM use cases in 6-8 weeks across different domains.
Co-creation, co-operation and generative AI @ Scale: Design and implementation services for prototyping and building effective business solutions (virtual assistants and knowledge hubs, for example) leveraging commercial or open source foundation models.
Bespoke foundation models: Leverage original innovations from IBM Research, AWS and other sources on foundation models for specialized domains (chemistry, material science and sensor data processing) to address bespoke domain specific use cases.
Foundation model fovernance, FMOps: Set up the required organizational and technical governance for scaling foundation models across the enterprise using IBM Consulting’s AI@Scale method.
Enterprises across industries are currently facing considerable pressure to adopt generative AI rapidly and demonstrate value. With more than 40K+ AI and analytics engagements worldwide, IBM Consulting has been consistently ranked as a leader by several analysts. IBM Consulting is committed to helping life sciences enterprises navigate and realize value from generative AI through the recently announced generative AI CoE, an immersive consultative process like IBM Garage and accelerators like ATOM. Clients need a trusted, experienced, and skillful partner to help them on their generative AI journey and IBM Consulting is ready to help them by meeting them where they are.